Elliott Wave

Elliott Wave describes price movement as a sequence of repeating, self-similar patterns rather than independent moves. Ralph Nelson Elliott set it out in the 1930s, and the claim is structural: markets advance in a five-wave sequence in the direction of the larger trend and correct in a three-wave sequence against it, and that shape repeats at every scale.

The five-wave move is labelled 1 through 5 and called an impulse; waves 1, 3 and 5 travel with the trend, waves 2 and 4 retrace against it. The correction is labelled A-B-C. Because the structure is fractal, each wave of a larger impulse decomposes into a smaller impulse or correction of its own, which is why practitioners talk about *degree*: the scale at which a count is read.

## How the mechanics work

Three conditions are rules, not tendencies: wave 2 never retraces more than 100% of wave 1; wave 3 is never the shortest of waves 1, 3 and 5; and wave 4 does not enter the price territory of wave 1, except in diagonal formations. Breaking any of them invalidates the count, which gives the framework a built-in invalidation level for risk placement.

Everything else is a guideline. Alternation holds that if wave 2 is a sharp correction, wave 4 tends to be sideways. Fibonacci proportion is the most used of all: wave 2 commonly retraces 50–61.8% of wave 1, wave 4 around 38.2% of wave 3, and wave 3 frequently extends to roughly 1.618 times wave 1. These are tendencies, not constraints.

Corrections have their own taxonomy — zigzags, flats, triangles and longer combinations — and matter for one reason: a correction ends where the next impulse begins, so reading which one is unfolding is how a trader positions ahead of wave 3 or wave 5.

## Main variants

**Discretionary counting** applies the full rule set and Fibonacci projections manually. **Oscillator proxies** replace counting with a momentum reading: the Elliott Wave Oscillator, the difference between a short and a long moving average of price, marks the momentum peak associated with wave 3 and the divergence associated with wave 5; it approximates the phase rather than labelling waves. **Automatic labelling indicators** build a count algorithmically from a swing detector such as ZigZag or fractals, then apply the rule checks. **Correction-focused versions** skip impulse trading and treat the completion of a zigzag or flat as the entry into the next leg. **Hybrid versions** use the count only as context and delegate the trigger to something mechanical: a structure break, a moving-average cross, a candle pattern at a Fibonacci level.

## What typically differentiates implementations

Two versions can both call themselves Elliott Wave and share almost nothing operationally. Swing detection comes first — the ZigZag percentage, depth or ATR multiple that defines a pivot determines the entire count. Degree and timeframe come second: a count on a 5-minute chart and one on a weekly chart are different strategies. After that come the details that decide the trades — whether relabelling is permitted as new bars arrive, which rules are enforced versus treated as soft, the Fibonacci tolerance allowed, where invalidation sits, whether targets come from extensions or fixed multiples of risk, and how alternate counts are handled.

## Common mistakes

Repainting is the dominant error. A wave label that changes after later bars print is not a signal, and a chart of past data with a clean count on it shows what the rules concluded at the end, not in real time. Backtesting on hindsight labels is the same error.

Beyond that: forcing a count onto price that does not fit instead of accepting that none is valid yet; treating Fibonacci guidelines as rules; mixing degrees, so a wave 4 at one scale is compared against a wave 2 at another; and reading an oscillator value as if it were a count. Sample size is a quieter problem — completed five-wave sequences at a given degree are infrequent, so a version tested on one instrument over a few years rests on few independent observations.

## How to evaluate and backtest a version

Start by removing discretion. Pivot detection, rule checks, ratio tolerances, entry trigger, invalidation and target must be specified precisely enough that two people coding them independently produce the same trades; if a step still requires judgment, the version is not testable as written.

Then test for lookahead. Replay the data bar by bar and log the label assigned at each close; if labels revise, measure how often and by how much, since that revision rate is the version's real-time uncertainty. Log invalidated counts alongside completed ones — the counts that broke a rule are the informative half of the sample.

Segment by what the count claims, since wave 3 and wave 5 entries are different trades. Run sensitivity checks on the swing threshold and Fibonacci tolerance; an edge that appears only at one ZigZag setting is fitted. Include costs and slippage, validate out of sample and across several instruments, and compare against a plain baseline such as a simple retracement entry, which the count has to beat to justify its complexity.

Three decoded versions are linked from this page: an oscillator-and-trend-indicator implementation (Inversiones En el Mundo), an indicator-driven identification version (The Good, The Bad And The Bitcoin), and a theory-led approach built around zigzag corrections (Elliott Wave Forecast). Each page states the rules and parameters exactly as presented in its source video.

Strategies in this concept (8)

Frequently asked questions

Is Elliott Wave objective enough to backtest?

Only once every step is specified. The three hard rules are objective, but the count itself depends on which swings you consider significant, and that choice is made by the trader or by a pivot detector with a threshold. A version becomes testable when pivot detection, rule enforcement, Fibonacci tolerances, entry trigger and invalidation are all written as fixed parameters. If any of them is left to interpretation, the results measure your interpretation, not the method.

What is the difference between the rules and the guidelines?

There are three rules: wave 2 cannot retrace more than 100% of wave 1, wave 3 cannot be the shortest of waves 1, 3 and 5, and wave 4 cannot overlap the price territory of wave 1 outside diagonal formations. Breaking any of them invalidates the count. Everything else — alternation, channelling, the usual Fibonacci proportions — is a tendency observed often enough to be useful for ranking candidate counts, but a count that ignores a guideline is still a valid count.

Does the Elliott Wave Oscillator actually count waves?

No. It is the difference between a short and a long moving average of price, so it measures momentum, not structure. It is used as a proxy: the momentum extreme it prints tends to coincide with the strongest part of an impulse, and a lower momentum reading on a higher price high is read as the divergence typical of a fifth wave. That approximates the wave phase, so a version built on it is a momentum strategy using Elliott vocabulary rather than an implementation of the count.

Why do automatic Elliott Wave indicators repaint, and does it matter?

They repaint because a pivot is only confirmed once price has moved far enough away from it, so the most recent labels are provisional and can be reassigned when new bars arrive. It matters a great deal for evaluation: a historical chart shows the final labelling, not what was visible at the time. The only reliable check is a bar-by-bar replay that records the label at each close and reports how often it later changed.

Which timeframe or degree should a count be built on?

The concept is scale-invariant in theory, so any timeframe can be counted, but the practical trade-off is real. Higher degrees produce fewer, longer sequences and therefore very small samples for testing; lower degrees produce many more counts but a higher proportion of ambiguous ones, since noise supplies plenty of swings that satisfy the rules by chance. Most implementations pick one degree for context and a lower one for timing, and it is worth checking that a version behaves consistently when that pairing shifts.

Do you need Fibonacci ratios to trade Elliott Wave?

Not strictly — the rules that define a valid count make no reference to Fibonacci. In practice most implementations use ratios for two jobs the rules do not cover: choosing entry levels inside a correction and projecting targets for the next impulse. Because they are guidelines rather than requirements, the tolerance allowed around each ratio is a parameter of the strategy and should be tested like any other, not adopted as a fixed number.

All strategy concepts · Explore all strategies